Chroma is a 8.9B parameter model based on FLUX.1-schnell (technical report coming soon!). It’s fully Apache 2.0 licensed, ensuring that anyone can use, modify, and build on top of it—no corporate gatekeeping. The model is still training right now. this model needs CFG so that's basically 2x slower compared to flux.
both flux schnell and dev are distilled model to make it faster, this model however is not distilled. The developer can distill it but is focusing on "re-pretraining" the missing concepts into the model. 1024x1024 (for now)
What Chroma Aims to Do
Training on a 5M dataset, curated from 20M samples including anime, furry, artistic stuff, and photos.
Fully uncensored, reintroducing missing anatomical concepts.
Built as a reliable open-source option for those who need it.
See the Progress
Hugging Face Debug Repo: https://huggingface.co/lodestones/chroma-debug-development-only
Live AIM Training Logs: https://training.lodestone-rock.com
Training code!: https://github.com/lodestone-rock/flow
CivitAi gallery: https://civitai.com/posts/13766416
CivitAi model: https://civitai.com/models/1330309/chroma
All credit goes to https://huggingface.co/lodestones/Chroma
Apache-2.0 license
I will update this tool when it's available
Will be testing/adding loras
it's training at 512px res so the details at 1024 wont be that good ATM. High res training is reserved at the very end of the training epochs (48-50)